Energy Efficient Time Synchronization in WSN

for Critical Infrastructure Monitoring

Aravinda S. Rao1, Jayavardhana Gubbi1, Tuan Ngo2,James Nguyen2, and Marimuthu Palaniswami1

1 ISSNIP, Dept. of Electrical and Electronic Engineering, The University ofMelbourne, Vic - 3010, Australia

{sridhara,jgl,palani}@unimelb.edu.au2 Infrastructure Protection Research Group, Dept. of Civil and Environmental

Engineering, The University of Melbourne, Vic - 3010, Australiadtngo@unimelb.edu.au, james.idt@gmail.com

Abstract. Wireless Sensor Networks (WSN) based Structural HealthMonitoring (SHM) is becoming popular in analyzing the life of criticalinfrastructure such as bridges on a continuous basis. For most of theapplications, data aggregation requires high sampling rate. A need foraccurate time synchronization in the order of 0.6−9 µs every few minutesis necessary for data collection and analysis. Two-stage energy-efficienttime synchronization is proposed in this paper. Firstly, the network is di-vided into clusters and a head node is elected using Low-Energy AdaptiveClustering Hierarchy based algorithm. Later, multiple packets of differentlengths are used to estimate the delay between the elected head and theentire network hierarchically at different levels. Algorithmic scheme lim-its error to 3-hop worst case synchronization error. Unlike earlier energy-efficient time synchronization schemes, the achieved results increase thelifetime of the network.

Keywords: Time Synchronization, Wireless Sensor Networks, CriticalInfrastructure Monitoring, StructuralHealth Monitoring, Energy Efficient.

1 Introduction

Time synchronization in WSN has been an important research area over thepast decade. Numerous protocols, benchmarked algorithms and approaches havebeen proposed to reduce the synchronization error and also being implementedto test their viaibility for WSN; Reference Broadcast Synchronization (RBS) [2],Timing-sync Protocol for Sensor Networks (TPSN) [3], Flooding Time Synchro-nization Protocol (FTSP) [10], Lightweight Time Synchronization for SensorNetworks (LTS) [4], Tiny-sync [15] and Mini-sync [15] are widely used.

Time synchronization is an inherent problem in any network. The time be-tween any two networked computers differs due to clock drift and clock offset.Often the crystal oscillator, provider of appropriate timing signals, is observedto be drifted from its normal specified frequency. The clock instability is caused

D.C. Wyld et al. (Eds.): NeCoM/WeST/WiMoN 2011, CCIS 197, pp. 314–323, 2011.c© Springer-Verlag Berlin Heidelberg 2011

Energy Efficient Time Synchronization in WSN 315

by many environmental factors, chiefly attributing to temperature variation [13].The frequency offset is the difference in clock timing between two clocks at anyparticular instant. As the frequency of the crystal varies, the clock offset variesi.e. offset increases or decreases relative to another clock.

Physical time plays a crucial role in WSN application for Critical Infrastruc-ture Monitoring (CIM) [1,8]. Multiple nodes possessing identical time-stampsof an event are essential. The significance of the physical time with respect toWSNs are detailed in [12] as : (a) interaction between a sensor network and theobserver, (b) interaction between nodes, and (c) interface between nodes and thereal world. Absence of real-time clock on WSN motes have given rise to softwareschemes to synchronize their on-board clocks. The Internet uses Network TimeProtocol (NTP) [11] having accuracy in the order of nano to pico seconds whichcan be adapted where there is no power constraint. Numerous solutions existin literature such as software clocks, unidirectional synchronization, round tripsynchronization and reference broadcasting having many advantages and disad-vantages; schemes based on high-rate data collection are shown to have decreasedfailure rate in synchronizing [12]. Power-constrained frequent synchronization isrequired to tackle drifting, a major contributor to error, with an accuracy of 0.6to 9 μs for modal analysis [9]. A 5 ms drift in 6 s period was reported by Wanget al. [17] which is on the higher side of acceptability.

In this paper, an energy-efficient time synchronization algorithm is proposedusing few existing algorithms, but combines them uniquely to achieve the desiredgoal. At first, it uses Low-Energy Adaptive Clustering Hierarchy (LEACH) fordividing the network into clusters based on nodes’ available energy [5]. Later,a new algorithm is proposed for synchronization. This paper is organized asfollows: Section 2 gives brief description of existing protocols, Section 3 containsthe proposed approach towards time synchronization followed by Section 4 withresults and discussion, and conclusions are given in Section 5.

2 Related Work

There has been substantial amount of work carried out in time synchronizationin general [16]. Most of the work has been done in early stage of WSN researchand not many of them have been tailored to specific application such as criti-cal infrastructure monitoring. As a result, energy has not been considered as acritical parameter while designing a time synchronization algorithm. As pointedout by Krishnamurthy et al. [9], there is a need to synchronize the clock morefrequently than originally intended and this will have a bearing on lifetime onthe network. In this context, a few time synchronization methods proposed inliterature have been identified to be very useful and are described below.

Reference Broadcast Synchronization (RBS), a pioneering work, is a receiver-to-receiver synchronization in the field. The main drawback of this approach is,as and when the number of receivers grow, the amount of message exchangeclimbs up [2] which is unsuitable for large networks. Flooding Time Synchro-nization Protocol (FTSP) floods the network with synchronization packets tothe neighbors [10]. However, FTSP cannot compensate the propagation delay.

316 A.S. Rao et al.

There have been a few other algorithms proposed similar to FTSP, however,this will not be very energy efficient [16]. Another very popular method onwhich a lot of methods have been developed is Timing-sync Protocol for SensorNetworks (TPSN) [3]. TPSN has two phases: firstly, during the level discoveryphase, the protocol forms a hierarchical network of different levels. Node withlevel 0 forms the top level followed by level 1, 2 etc. Node in the level i commu-nicates with at least a node in the level above (i-1 ) it. During the second phase(also called as synchronization phase), the node at level 0 (root node) initiatesthe synchronization phase using constrained flooding [16,3]. In case of messagetransmission from a node of higher level, though a node at non-immediate leveloverhears the message from higher levels, it only synchronizes to the level aboveit. Lightweight Time Synchronization for Sensor Networks (LTS) uses three mes-sages to synchronize a pair of nodes [4] at the pairwise synchronization stage. Itis extended from single-hop synchronization to centralized multi-hop network.The re-synchronization rate in LTS is dependent on the depth of the tree. Also,a bounded-clock drift is assumed and the drifts of all the nodes are consideredas equal. Although they discuss distributed multi-hop, energy efficiency is notaccomplished as the algorithm focussed only on time synchronization. Shahzadet al. [14] propose EETS algorithm which combines RBS and TPSN for achiev-ing better energy efficiency. Although this improves energy efficiency, in largenetworks with multi-hops reaching up to 8-10 levels, the synchronization error ishigher than required for critical infrastructure monitoring. More recently, Kim etal. [7] have developed an algorithm by reducing the number of packets exchangedthereby conserving power. This again is a combination of RBS, TPSN and LTSmethods but reduces the number of packets exchanged. Again, the topologyused is identical to TPSN and the best achievable error rate is affected by thenumber of hops. The second stage of the proposed algorithm uses the aboveclass of algorithms which is a combination of hierarchical levels as TPSN andpairwise synchronization phase of LTS in synchronizing the time after energyefficient clusters are created in the first phase using LEACH based approach.The approach is described in the following section.

3 Approach

Energy efficient time synchronization is carried out in two stages: (a) dividingthe network into clusters in an energy-efficient manner using LEACH [5] and(b) synchronizing the time within the cluster using multi-level pairwise synchro-nization. LEACH elects a cluster head periodically based on the highest energyavailable live node. The chance of becoming a cluster head is rotated randomlyensuring that energy from a single node is not drained [5]. Upon cluster head elec-tion, pairwise transmission range based multilevel, hierarchical sub-clusters areformed in the immediate vicinity of the parent node. Variable-length packets areused to determine the nondeterministic latency between sender and the receiverand is compensated before transmission. The algorithm ensures that the time ofa child node is same as that of the its parent. The elected four cluster heads andthe cluster formation is pictorially represented in Figure 1. According to LEACH

Energy Efficient Time Synchronization in WSN 317

Fig. 1. Schematic of cluster formation using LEACH. Numbers within the node repre-sent levels. B is the chosen cluster-head in any given round.

algorithm [5], the optimal number of nodes to be cluster heads was chosen to beN = 5%. In case of the proposed algorithm, as the number of nodes in any clus-ter is limited to 12 (empirically chosen), N = (Total Number of Nodes)/12 + 1is chosen for CIM. The second part of time synchronizing procedure is dividedinto four phases as explained in the following subsections.

3.1 Initiation

Let there be ’N’ nodes neighboring to base station node. Out of ’N’ nodes,only ’P’ of them are within the communication range of base station and theremaining (N-P) nodes are not. All nodes in the network are switched ON beforeany communication or network activity is initiated. Node B initiates its routineby synchronizing its time with the base station i.e. with the computer ensuringdatabase connection. It should be noted that the nodes in the topology are placedsuch that it is in the communication range of at least one node.

3.2 Registration

The process of adding a new node to the topology is called as ‘Registration’. Eachnode has its own ID and is uniquely identifiable. For discussion, let us considerN = 2. Nodes in the transmission range of B are therefore 1 and 2 and vice-versa. Therefore, the transmission power for both parent (B) and child nodes(1 and 2) are the same. After initiation process, B broadcasts a READY signalto inform its neighbors that it is ready to accept REQUEST from child nodes.Now, upon hearing B’s broadcast signal, both 1 and 2 will send a REQUESTsignals. When a READY signal is sent, B will wait for tw seconds to hear fromthe nodes. Any message received after tw is discarded. Suppose, B received both

318 A.S. Rao et al.

the requests from 1 and 2, then B will serve first node 1 followed by 2 in a TimeDivision Multiple Access (TDMA) slotted fashion. Nodes 1 and 2 are waitingfor the COMMAND signal from B. Nodes 1 and 2 will not proceed further untilthey receive any signal from B.

3.3 Calculating Nondeterministic Latency

Upon receiving and processing the COMMAND signal from B, node 1 will waitfor the tstart. At tstart, it will send message M1 to B. It is repeated for n times atevery trepeat intervals. Packets are time stamped at physical layer of node 1 andare stored in B as and when received. Next, B will send a COMMAND signal withM2 > M1 i.e. the message length greater than the previous one. At this time, Brecords node 1’s data. Let δtn be the time difference between consecutive packetsarrived at the base station from a particular node. Let LM1 be the messagelength of M1 and the time stamp of the packets’ arrival be t1...tn. Time differencebetween packets can be calculated as in equation 1, where n is the packet arrivedat time tn. B calculates the average time dM1 required by a packet of size M1 toreach base station due to send time, access time, propagation time and receivetime and is given by equation 2. Let LM2 be the length of M2. Now, B sendsa second COMMAND signal with M2 > M1. Equations 1 and 3 are calculatedfor M2. Then essentially, the time required for packet of size LM2 − LM1 to beconstructed, transferred and received from node 1 (or node 2) to B is given byequation 4. Now B sends tByte in a REGISTRATION COMPLETE message tonode 1. B accepts packets strictly from level 1, and level 1 nodes accept data fromlevel 2 only, identical to TPSN [3]. After this, node 1 sends a READY commandto its neighbors. At the same time, B sends a COMMAND signal to node 2 andthe process continues. At the end of REGISTRATION COMPLETE with 2, Bsends an ACCEPT command to accept data from 1 and 2. Node 2 continuesto act as parent and starts broadcasting READY signal to its neighbors (childnodes).

δtn = |tn − tn−1| (1)

dM1 =∑n

i=1 δtnn

(2)

dM2 =∑n

i=1 δtnn

(3)

tbyte =(dM2 − dM1)(LM2 − LM1)

(4)

3.4 Time Synchronization

After every T seconds predetermined by application and the frequency of datacollection, B multi-casts current time to all the registered nodes. T is calculatedusing equation 5, where ti is the time required for a ith node to be registered,

Energy Efficient Time Synchronization in WSN 319

taccept is the time for accepting data from all the registered nodes. The typicaltree structure formed during time synchronization after selecting the cluster headis illustrated in Figure 2.

T =∑N

i=1 tiN

+ taccept (5)

The term∑ N

i=1 ti

N is required if a node child node fails and wants to re-register.

Fig. 2. Hierarchical level creation during proposed second stage of time synchronization

In (5), first part is ’Registration time’ and the second part is ’Data acceptancetime’ in the proposed scheme. Additionally, if there is a new node, then it canregister itself in the first part and send data subsequently. In general, T shouldbe selected such that allowances are made for any new node to be added.

3.5 Packets Required to Synchronize a Node

Determining number of packets ’n’ in the COMMAND phase is a two-step pro-cess. During the first step, B sends COMMAND with n=5. The maximum toler-able synchronization error (E) between two nodes is specified by the user as tE .During the second step, the node calculates the δEn and the average of δEn asin equations 6 and 7 respectively. If δEavg is more than predetermined value δtE ,then B raises its bar and requests more packets during M2 acquisition, otherwiseM2 will have ‘n’ packets as well; B tries to find the delay due to variable-lengthpacket size.

δEn = δtn − trepeat (6)

δEavg =∑n

i=1 δEn

n(7)

4 Results and Discussion

The proposed energy-efficient time synchronization algorithm can be catego-rized as peer-to-peer, externally synchronized, deterministic, sender-to-senderand clock-corrected approach analogous with table 2 of [16]. It gives an added

320 A.S. Rao et al.

advantage of selectively using the node head for energy efficiency. Having pre-sented the simulation results in this paper, in future, we plan to validate theproposed algorithm on iMote2 platform and the implementation is underway.At the end of this section, theoretical energy consumption and effect of packetlength with respect to iMote2 platform are presented.

LEACH is a very popular clustering-based routing protocol, has been shownto minimize global energy usage by distributing the load among nodes [5]. The-oretically and experimentally, it has been shown that the lifetime of the networkis increased 3 times compared to a static topology. In the first stage of the pro-posed time synchronization, LEACH based clustering is used for dividing thenetwork into clusters. From the results presented in [2,4,3,10], it is very clearthat the clock error increases as the number of hops increase. Using cluster-based step at the first stage, ensures that the number of hops required in thesecond stage of the algorithm will never cross three levels (12 nodes). Hence, themaximum error in time synchronization by the proposed algorithm is the worstcase 3-hop error that of TPSN and LTS. According to Krishnamurthy et al. [9],anything less than 10 μs error for 30 minutes duration is acceptable for modalanalysis using acceleration sensors. By virtue, the proposed algorithm ensuresthis criterion. Moreover, the clustering algorithm enables data aggregation usingenergy-efficient routing apart from time synchronization.

4.1 Simulation Analysis

The experiment was conducted using OMNeT++, a component-based discreteevent network simulation package. OMNeT++ offer tools to simulate computernetworks, queuing networks, processor architectures and for distributed systems.We used NICTA’s Castalia 3.1 framework on top of OMNeT++ 4.1 to simulateWSN scenarios [6]. In particular, for this work, we developed three simulationscenarios with three nodes - one base base station node (N0) and two sink nodes(N1, N2). N0 is the head of the nodes and the time-provider for other twonodes. The experiment was simulated for 10 s, with synchronization period of1 s each for a total period of 10 s. Start up delay for the nodes were initializedto 0.5 ms. Node N0 waited for N1 and N2 for their reply after sending the ini-tiation (READY) signal. The waiting time (tw) was kept at 400 ms followingthe broadcast of READY signal to allow sufficient time to respond. Both N1and N2 nodes replied with REQUEST signal. After tW seconds, N0 processedREQUEST in ascending order of the node IDs. Node N0 sent a COMMANDsignal to N1 with tstart=10 ms, trepeat=10 ms, n=5, M1=5, M2=15. N1 sentmessages to N0 after adding current time with tstart at trepeat intervals (M1=5,M2=15 bytes of data sent ’n’ times). Node N0 calculated δtn for M1 and M2separately and then the average of δtn i.e. dM1 and dM2. tByte was calculatedusing equation 4. Node N0 calculated tByte and δEavg, and sent the tByte toN1 as variable-length packet delay with REGISTRATION COMPLETE signal.After processing nodes N1 by N0, N2 was processed by sending COMMANDsignal. Further, after processing N2, node N0 sent ACCEPT signal at start oftaccept. In our simulation taccept=0 s as only two sink nodes were considered. At

Energy Efficient Time Synchronization in WSN 321

T = 1 s from the start of the READY signal, N0 multi-casts current time to reg-istered nodes. Following this, N0 sent READY signal for the next round of clocksynchronization and this process of synchronization was continued at differentlevels. We simulated for only one level i.e. between N0 and N1-N2. In this sim-ulation for each time synchronization cycle, nodes N0 and N1 sent REQUESTsand participated in registration; however, during implementation, registrationwill be done once only. Media Access Control (MAC) layer was bypassed in thesimulation environment; radio was in Ideal mode with a transmission power levelof 0 dBm and configured to zero interference model; wireless channel was set tohave bidirectionally identical signal quality links between nodes. One-hop rout-ing was carried out by application layer. The simulation results are presentedfor three different scenarios of node placements in order to analyze the efficiencyof the proposed algorithm.

• Scenario 1 - Equidistant nodes: N1 and N2 with distance equal to 14.14m from N0 and trepeat = 10 ms; for Node N1 tByte was 8.27 ms, δEavg

= 67.989, and tByte = 20.65 ms for N2, δEavg =33.933 was obtained bysimulation. Using this, nodes N1 and N2 sent 10 messages every 1 s forten seconds. The message lengths were 5 and 15 bytes. For the two nodes,the results are summarized in table 1 for seven rounds. The nondeterministicdelay associated with size of a packet can be calculated as a correction factor(tByte∗ Number Of Bytes) and is applied before packet transmission.

Table 1. Synchronization Errors for Node 1 and Node 2

Round Node Sync Error Node Sync ErrorID (ms) ID (ms)

1 1 41.39 2 99.432 1 29.82 2 95.553 1 41.34 2 91.724 1 37.50 2 91.725 1 37.52 2 95.566 1 41.34 2 95.567 1 37.52 2 95.56

• Scenario 2 - Unequal distance between nodes: Distance between N0and N1 was 14.14 m, and between N0 and N2 was 26.92 m; the resultsobtained from this simulation were same as Scenario 1, with distance havingless impact on nondeterministic latency. This can also be justified by the factthe travel times for radio waves remains nearly unvaried for short distances.

• Scenario 3 - Synchronization error at different levels: Suppose leveli is the base station level, level i + n is the node at level n, then the error atdifferent levels can be calculated relative to base station level as equation 8,where tByte(i+n)−i is the time required from a byte to be transferred betweennodes of level n and i; Ni is the number of bytes to be transferred. The errorat level i + n is the cumulative error from level i + n to i.

322 A.S. Rao et al.

Errorni =i∑

n

tByte(i+n)−i ∗ Ni (8)

4.2 Energy Consumption

Energy consumed by a node in general is essentially due to two main contributors- Processor and the Radio. During active times, the available energy is utilized(by processor and/or RI) up to or less than their stated absolute maximumrating values. During sleep (or idle) times, the nodes conserve their energy byentering into the low-power modes, in particular, sleep modes deactivate RI.Energy consumed by a node for the purpose of the communication, in general,at any particular instant, is given by equation 9; energy consumed by the mainprocessor of the node is given by equation 10 and the total energy consumed bya node (at any instant of time) is given by equation 11

ERI = EReceiver + ESender + ERadio Process (9)

EP = EProcess ∗ [TRadioON + TRadioOFF ] + ESleep ∗ TSleep (10)

EN = EProcessor + ERI (11)

From equations 9 and 11, for Imote2, EPmax = 97 mA; EPmin = 31.387 mA;ERImax = 36.626 mA; ERImin=27.5 mA. Energy expended in sending (Esend)and receiving (EReceive) a packet are 31 mA and 18.1 mA respectively. For thecluster shown in Figure 2, energy consumed will be 32 w per cycle per cluster. Forclusters with different spatial orientation, energy consumption entirely dependson number of clusters in the vicinity of each cluster head.

5 Conclusion

Time synchronization is fundamental to data fusion and henceforth to help detectand/or monitor events. The accuracy of synchronized time among nodes, cou-pling closely (in time), ameliorates the systems’s ability to determine an event.Synchronization of time among nodes can be achieved by eliminating nondeter-ministic delays (send time, access time, propagation time, receive time). Addingto nondeterministic delays, frequency drifts due to clock instabilities and off-set due to drifts, contribute to increase in synchronization error or mismatchin clock timings among nodes. Though several algorithms and approaches havebeen proposed by researchers for time synchronization, this proposed approach isalso equally useful for energy-efficient, robust, externally synchronized multi-hopnetworks.

Energy Efficient Time Synchronization in WSN 323

References

1. Elson, J.: Time Synchronization in Wireless Sensor Networks. Ph.D. thesis, UCLA(2003)

2. Elson, J., Girod, L., Estrin, D.: Fine-grained network time synchronization usingreference broadcasts. In: Proceedings of the 5ht Symposium on Operating SystemsDesign and Implementation (OSDI 2002), vol. 36, pp. 11–19. ACM, New York (2002)

3. Ganeriwal, S., Kumar, R., Srivastava, M.B.: Timing-sync protocol for sensor net-works. In: Proceedings of the 1st International Conference on Embedded NetworkedSensor Systems (SenSys 2003), pp. 138–149. ACM, New York (2003)

4. Greunen, J.V., Rabaey, J.: Lightweight time synchronization for sensor networks.In: Proceedings of the 2nd ACM International Conference on Wireless Sensor Net-works and Applications (WSNA 2003), pp. 11–19. ACM, New York (2003)

5. Heinzelman, W.R., Chandrakasan, A., Balakrishnan, H.: Energy-efficient commu-nication protocol for wireless microsensor networks. In: Proceedings of the 33rd An-nual Hawaii International Conference on System Sciences, vol. 2, pp. 1–10. IEEE,Los Alamitos (2000)

6. Castalia, http://castalia.npc.nicta.com.au/ (January 2011 (verified on 6thFebruary 2011))

7. Kim, B.K., Hong, S.H., Hur, K., Eom, D.S.: Energy-efficient and rapid time syn-chronization for wireless sensor networks. IEEE Transactions on Consumer Elec-tronics 56(4), 2258–2266 (2010)

8. Kim, S.: Wireless Sensor Networks for Structural Health Monitoring. Master’sthesis, UC-Berkely (2005)

9. Krishnamurthy, V., Fowler, F., Sazonov, E.: The effect of time synchronisation ofwireless sensors on the modal analysis of structures. Smart Materials and Struc-tures 17 (August 2008)

10. Maroti, M., Kusy, B., Simon, G., Ledeczi, A.: The flooding time synchronizationprotocol. In: Proceedings of the 2nd International Conference on Embedded Net-worked Sensor Systems (SenSys 2004), pp. 138–149. ACM, New York (2004)

11. Mills, D.L.: Internet time synchronization: The network time protocol. IEEE Trans-actions on Communications 39, 1482–1493 (1991)

12. Rmer, K., Blum, P., Meier, L.: Time Synchronization and Calibration in WirelessSensor Networks. In: Handbook of Sensor Networks: Algorithms and Architectures,pp. 199–237. Wiley and Sons, Chichester (2005)

13. Schmid, T., Charbiwala, Z., Friedman, J., Cho, Y.H., Srivastava, M.B.: Exploitingmanufacturing variations for compensating environment-induced clock drift in timesynchronization. In: Proceedings of the 2008 ACM SIGMETRICS InternationalConference on Measurement and Modeling of Computer Systems (SIGMETRICS2008), pp. 97–108. ACM, New York (2008)

14. Shahzad, K., Ali, A., Gohar, N.: Etsp: An energy-efficient time synchronizationprotocol for wireless sensor networks. In: 22nd International Conference on Ad-vanced Information Networking and Applications - Workshops (Aina Workshops2008), pp. 971–976 (2008)

15. Sichitiu, M.L., Veerarittiphan, C.: Simple, accurate time synchronization for wire-less sensor networks. In: IEEE Wireless Communications and Networks (WCNC2003), vol. 2, pp. 1266–1273 (2003)

16. Sundararaman, B., Buy, U., Kshemkayani, A.D.: Clock synchronization for wirelesssensor networks: A survey. Ad Hoc Networks 3, 281–323 (2005)

17. Wang, Y., Lynch, J.P., Kae, K.H.: A wireless structural health monitoring systemwith multithreaded sensing devices. Struct. Infrastruct. Eng. 3, 103–120 (2007)